Importing relevant libraries

import numpy as np
import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.metrics import accuracy_score, f1_score, confusion_matrix, auc, roc_curve, classification_report
from sklearn.preprocessing import StandardScaler
import matplotlib.pyplot as plt
import seaborn as sns
import plotly.express as px
from sklearn.preprocessing import LabelEncoder
from sklearn.ensemble import RandomForestClassifier

Data Ingestion

dt_train = pd.read_csv('ecommerce_train.csv')
test = pd.read_csv('ecommerce_test.csv')

dt_train.head()

	session_id	administrative	administrative_duration	informational	informational_duration	productrelated	productrelated_duration	bouncerates	exitrates	pagevalues	specialday	month	operatingsystems	browser	region	traffictype	visitortype	weekend	revenue
0	139957	2	47.500000	0	0.0	19	1203.480714	0.000000	0.027421	6.748509	0.0	Nov	2	2	3	2	Returning_Visitor	False	0
1	496587	0	0.000000	0	0.0	7	152.000000	0.000000	0.033333	0.000000	0.0	Dec	3	2	6	3	Returning_Visitor	False	0
2	440268	7	206.500000	1	389.9	95	4415.889952	0.000133	0.033750	2.676062	0.0	Nov	2	2	1	20	Returning_Visitor	False	0
3	294672	1	18.258571	0	0.0	103	8305.048706	0.004871	0.034746	6.824908	0.0	Nov	2	2	1	2	Returning_Visitor	True	1
4	497475	0	0.000000	0	0.0	42	1305.108333	0.016068	0.039742	0.000000	0.0	Dec	3	2	5	2	Returning_Visitor	False	0

test.head(2)

	session_id	administrative	administrative_duration	informational	informational_duration	productrelated	productrelated_duration	bouncerates	exitrates	pagevalues	specialday	month	operatingsystems	browser	region	traffictype	visitortype	weekend
0	283149	0	0.0	0	0.0	10	388.0	0.000000	0.014286	0.0	0.0	Nov	1	2	8	2	New_Visitor	False
1	253020	0	0.0	0	0.0	17	142.4	0.017707	0.044623	0.0	0.0	Nov	1	2	3	8	New_Visitor	False

dt_train.shape

(40000, 19)

test.shape

(10000, 18)

dt_train.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 40000 entries, 0 to 39999
Data columns (total 19 columns):
 #   Column                   Non-Null Count  Dtype  
---  ------                   --------------  -----  
 0   session_id               40000 non-null  int64  
 1   administrative           40000 non-null  int64  
 2   administrative_duration  40000 non-null  float64
 3   informational            40000 non-null  int64  
 4   informational_duration   40000 non-null  float64
 5   productrelated           40000 non-null  int64  
 6   productrelated_duration  40000 non-null  float64
 7   bouncerates              40000 non-null  float64
 8   exitrates                40000 non-null  float64
 9   pagevalues               40000 non-null  float64
 10  specialday               40000 non-null  float64
 11  month                    40000 non-null  object 
 12  operatingsystems         40000 non-null  object 
 13  browser                  40000 non-null  object 
 14  region                   40000 non-null  int64  
 15  traffictype              40000 non-null  object 
 16  visitortype              40000 non-null  object 
 17  weekend                  40000 non-null  bool   
 18  revenue                  40000 non-null  int64  
dtypes: bool(1), float64(7), int64(6), object(5)
memory usage: 5.5+ MB

dt_train.isna().sum()

session_id                 0
administrative             0
administrative_duration    0
informational              0
informational_duration     0
productrelated             0
productrelated_duration    0
bouncerates                0
exitrates                  0
pagevalues                 0
specialday                 0
month                      0
operatingsystems           0
browser                    0
region                     0
traffictype                0
visitortype                0
weekend                    0
revenue                    0
dtype: int64

dt_train.describe()

	session_id	administrative	administrative_duration	informational	informational_duration	productrelated	productrelated_duration	bouncerates	exitrates	pagevalues	specialday	region	revenue
count	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000	40000.000000
mean	311469.372075	2.544150	89.794383	0.555225	40.977236	34.026975	1148.105672	0.021969	0.042307	6.494963	0.063553	3.241975	0.158275
std	108498.432497	3.667252	182.964645	1.332627	153.114387	46.798102	1666.181462	0.048533	0.048633	17.710646	0.199463	2.423586	0.365003
min	123460.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	1.000000	0.000000
25%	217950.000000	0.000000	0.000000	0.000000	0.000000	7.000000	183.883333	0.000000	0.013389	0.000000	0.000000	1.000000	0.000000
50%	311194.500000	1.000000	11.200000	0.000000	0.000000	19.000000	560.075000	0.003304	0.025000	0.000000	0.000000	3.000000	0.000000
75%	404680.500000	4.000000	110.154167	0.000000	0.000000	41.000000	1420.536667	0.016667	0.050000	0.000000	0.000000	4.000000	0.000000
max	499982.000000	22.000000	1946.000000	11.000000	1767.666667	440.000000	18504.126210	0.200000	0.200000	226.677702	1.000000	9.000000	1.000000

dt_train.describe(include='O')

	month	operatingsystems	browser	traffictype	visitortype
count	40000	40000	40000	40000	40000
unique	10	6	11	15	3
top	May	2	2	2	Returning_Visitor
freq	11048	21638	25540	12545	34496

dt_train.select_dtypes("object").nunique()

month               10
operatingsystems     6
browser             11
traffictype         15
visitortype          3
dtype: int64

categorical_features = dt_train.select_dtypes("object").columns
len(categorical_features)

5

for col in categorical_features:
    print(dt_train[col].value_counts())

May     11048
Nov      9625
Mar      6086
Dec      5621
Oct      1661
Sep      1503
Aug      1474
Jul      1429
June      857
Feb       696
Name: month, dtype: int64
2    21638
3     8439
1     8104
4     1498
8      236
*       85
Name: operatingsystems, dtype: int64
2     25540
1      7557
4      2783
5      1723
6       616
10      564
3       412
8       392
13      198
7       165
*        50
Name: browser, dtype: int64
2     12545
1      7323
3      6753
4      3622
13     2860
10     1510
6      1457
8      1270
5       874
11      728
20      563
15      147
9       124
*       120
7       104
Name: traffictype, dtype: int64
Returning_Visitor    34496
New_Visitor           5224
Other                  280
Name: visitortype, dtype: int64

Checking for target column distribution

cnt = dt_train['revenue'].value_counts()
fig,ax = plt.subplots(figsize=(5,5))
ax.pie(x=cnt.values,labels=cnt.index, autopct='%1.1f%%',colors=['cyan','yellow'],explode=(0,0.1))
ax.set_title('Revenue Percentage');

As we can see from above countplot, the dataset is highly imbalanced

EDA

rev_month = dt_train['month'].value_counts()
months = pd.DataFrame(data=rev_month.index, columns=["month"])
months['values'] = rev_month.values

fig = px.pie(months, values='values', names='month', color_discrete_sequence=px.colors.sequential.RdBu, title= "Counts of months")
fig.show()

def histogram(data,path,color,title,xaxis,yaxis):
    fig = px.histogram(data, x=path,color=color)
    fig.update_layout(
        title_text=title,
        xaxis_title_text=xaxis, 
        yaxis_title_text=yaxis, 
        bargap=0.2, 
        bargroupgap=0.1
    )
    fig.show()

calendar_months = ['Jan', 'Feb', 'Mar', 'Apr','May','June', 'Jul', 'Aug','Sep', 'Oct', 'Nov', 'Dec']

def histogram_1(data,path,color,title,xaxis,yaxis):
    fig = px.histogram(data, x=path,color=color,category_orders=dict(month=calendar_months))
    fig.update_layout(
        title_text=title,
        xaxis_title_text=xaxis, 
        yaxis_title_text=yaxis, 
        bargap=0.2, 
        bargroupgap=0.1
    )
    fig.show()

histogram_1(dt_train,"month","revenue",'Monthly revenue','month','Count')

Nov, May, Dec and March generating more revenues than other months

rev_region = dt_train['region'].value_counts()
region = pd.DataFrame(data=rev_region.index, columns=["region"])
region['values'] = rev_region.values

fig = px.pie(region, values='values', names='region', color_discrete_sequence=px.colors.sequential.RdBu, title= "Counts of regions")
fig.show()

histogram(dt_train,"region","revenue",'Region wise revenue','region','Count')

Region 1 has highest revenue

rev_os = dt_train['operatingsystems'].value_counts()
os = pd.DataFrame(data=rev_os.index, columns=["operatingsystems"])
os['values'] = rev_os.values

fig = px.pie(os, values='values', names='operatingsystems', color_discrete_sequence=px.colors.sequential.RdBu, title="Counts of operating systems")
fig.show()

histogram(dt_train,"operatingsystems","revenue",'Revenue on operatingsystems','operatingsystems','Count')

Operating system 2 has the maximum revenue

rev_browser = dt_train['browser'].value_counts()
browser = pd.DataFrame(data=rev_browser.index, columns=["browser"])
browser['values'] = rev_browser.values

fig = px.pie(browser, values='values', names='browser', color_discrete_sequence=px.colors.sequential.RdBu, title= "Counts of Browser")
fig.show()

histogram(dt_train,"browser","revenue",'Revenue on browser','browser','Count')

Browser 2 has maximum revenue

rev_traffictype = dt_train['traffictype'].value_counts()
traffictype = pd.DataFrame(data=rev_traffictype.index, columns=["traffictype"])
traffictype['values'] = rev_traffictype.values

fig = px.pie(traffictype, values='values', names='traffictype', color_discrete_sequence=px.colors.sequential.RdBu,title="Counts of traffictype")
fig.show()

histogram(dt_train,"traffictype","revenue",'Revenue on traffictype','traffictype','Count')

Traffictype 2 has highest revenue

rev_visitortype = dt_train['visitortype'].value_counts()
visitortype = pd.DataFrame(data=rev_visitortype.index, columns=["visitortype"])
visitortype['values'] = rev_visitortype.values

fig = px.pie(visitortype, values='values', names='visitortype', color_discrete_sequence=px.colors.sequential.RdBu, title= "Counts of visitors")
fig.show()

histogram(dt_train,"visitortype","revenue",'Revenue on visitortype','visitortype','Count')

Returning_Visitor has maximum revenue

histogram(dt_train,"weekend","revenue",'Revenue on weekends','weekend','Count')

Revenue on Weekend is maximum

px.scatter(dt_train, x='administrative', y='bouncerates',color='revenue')

px.scatter(dt_train, x='administrative', y='exitrates',color='revenue')

px.scatter(dt_train, x='administrative', y='pagevalues',color='revenue')

px.scatter(dt_train, x='informational', y='bouncerates',color='revenue')

px.scatter(dt_train, x='informational', y='exitrates',color='revenue')

px.scatter(dt_train, x='informational', y='pagevalues',color='revenue')

px.scatter(dt_train, x='productrelated', y='bouncerates',color='revenue')

px.scatter(dt_train, x='productrelated', y='exitrates',color='revenue')

px.scatter(dt_train, x='productrelated', y='pagevalues',color='revenue')

Observation

• when exit rate/bouncerate increases, the probability of revenue becoming '0' is very hich and when pagevalue increases the probability of revenue becoming '1' is high.

fig = px.scatter(dt_train, 
                 x = 'productrelated' , 
                 y = 'productrelated_duration', 
                 color='revenue')
fig.show()

Product categories from 0 to 200 are most viewed

Outliers Detection

sns.boxplot(x='administrative_duration',data=dt_train);

sns.boxplot(x='informational_duration',data=dt_train);

sns.boxplot(x='productrelated_duration',data=dt_train);

sns.boxplot(x='bouncerates',data=dt_train);

sns.boxplot(x='exitrates',data= dt_train);

sns.boxplot(x='pagevalues',data=dt_train);

sns.boxplot(x='specialday',data=dt_train);

sns.boxplot(x='region',data= dt_train);

Observation

• It seems like this data set resembles real time data, it has time as feature; in such a case a customer can spend as long time as they wish and that is considered normal. Hence it is not suitable to consider these extreme values as outliers. Hence we do not remove them.

Correlation Analysis

plt.figure(figsize=(12,12))
sns.heatmap(dt_train.corr(),annot=True,cmap='coolwarm');

session_ids_for_test = test.session_id

In the following we are dropping the session id and keep only one of the highly correlated columns by looking at the heatmap. What we find is that these columns do not impact the accuracy score. The score does not change either with or without keeping the columns in the dataset.

dt_train.drop(["session_id","productrelated_duration","exitrates","administrative_duration","informational_duration"],axis=1,inplace=True)

test.drop(["session_id","productrelated_duration","exitrates","administrative_duration","informational_duration"],axis=1,inplace=True)

dt_train['revenue'].value_counts()

0    33669
1     6331
Name: revenue, dtype: int64

X = dt_train.drop('revenue', axis = 1)
y = dt_train.revenue

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

for col in categorical_features:
    le_model = LabelEncoder()
    X_train[col] = le_model.fit_transform(X_train[col])
    X_test[col] = le_model.transform(X_test[col])
    test[col] = le_model.transform(test[col])

scaled_model = StandardScaler()
X_train = scaled_model.fit_transform(X_train)
X_test = scaled_model.transform(X_test)
test = scaled_model.transform(test)

Support Vector Classifier

from sklearn.svm import SVC

svc = SVC()

svc.fit(X_train, y_train)

SVC()

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

print(classification_report(y_train, svc.predict(X_train)))

              precision    recall  f1-score   support

           0       0.92      0.97      0.94     26915
           1       0.75      0.54      0.63      5085

    accuracy                           0.90     32000
   macro avg       0.83      0.75      0.78     32000
weighted avg       0.89      0.90      0.89     32000

print(classification_report(y_test,svc.predict(X_test)))

              precision    recall  f1-score   support

           0       0.92      0.96      0.94      6754
           1       0.72      0.52      0.60      1246

    accuracy                           0.89      8000
   macro avg       0.82      0.74      0.77      8000
weighted avg       0.89      0.89      0.89      8000

Visualising confusion matrix

from scikitplot.metrics import plot_confusion_matrix

plot_confusion_matrix(y_train, svc.predict(X_train));

plot_confusion_matrix(y_test, svc.predict(X_test));

Decision tree with hyperparameter tuning and crossvalidation

dtc= DecisionTreeClassifier()

params_dtc = {'max_depth':list(range(2,20))} # specifying the range for max_depths to create multiple trees

gs_dtc = GridSearchCV(estimator=dtc, param_grid=params_dtc, 
                      scoring='f1', cv=5).fit(X_train, y_train)
##### Manual
gs_dtc.best_params_, gs_dtc.best_score_

({'max_depth': 7}, 0.6593116445509427)

print(classification_report(y_train, gs_dtc.best_estimator_.predict(X_train)))

              precision    recall  f1-score   support

           0       0.93      0.96      0.94     26915
           1       0.74      0.62      0.67      5085

    accuracy                           0.90     32000
   macro avg       0.83      0.79      0.81     32000
weighted avg       0.90      0.90      0.90     32000

print(classification_report(y_test,
                            gs_dtc.best_estimator_.predict(X_test)))

              precision    recall  f1-score   support

           0       0.93      0.95      0.94      6754
           1       0.71      0.62      0.66      1246

    accuracy                           0.90      8000
   macro avg       0.82      0.79      0.80      8000
weighted avg       0.90      0.90      0.90      8000

Visualising confusion matrix

from scikitplot.metrics import plot_confusion_matrix

plot_confusion_matrix(y_train, gs_dtc.best_estimator_.predict(X_train));

plot_confusion_matrix(y_test, gs_dtc.best_estimator_.predict(X_test));

Visualising ROC_AUC curve

from scikitplot.metrics import plot_roc
y_pred_test = gs_dtc.best_estimator_.predict_proba(X_test)
plot_roc(y_test,y_pred_test);

Visualising the fitted Decision tree classifier

from sklearn.tree import plot_tree

_, ax = plt.subplots(nrows=1, ncols=1, figsize=(19,15), dpi = 300) # Resize figure
plot_tree(gs_dtc.best_estimator_, filled=True, ax=ax, fontsize=7)
ax.set_title("Decision tree")
plt.show()

Random Forest with hyperparameter tuning with crossvalidation

rfc= RandomForestClassifier()

params_rfc = {'max_depth':list(range(2,20)),'n_estimators':[10,25,50,100,200,500] } # specifying the range for max_depths to create multiple trees

gs_rfc = GridSearchCV(estimator=rfc, param_grid=params_rfc, scoring='f1', cv=5).fit(X_train, y_train)
##### Manual
gs_rfc.best_params_, gs_rfc.best_score_

({'max_depth': 12, 'n_estimators': 500}, 0.669963012788441)

print(classification_report(y_train, gs_rfc.best_estimator_.predict(X_train)))

              precision    recall  f1-score   support

           0       0.95      0.98      0.97     26915
           1       0.86      0.76      0.81      5085

    accuracy                           0.94     32000
   macro avg       0.91      0.87      0.89     32000
weighted avg       0.94      0.94      0.94     32000

Visualising confusion matrix

plot_confusion_matrix(y_train, gs_rfc.best_estimator_.predict(X_train));

print(classification_report(y_test, gs_rfc.best_estimator_.predict(X_test)))

              precision    recall  f1-score   support

           0       0.94      0.95      0.95      6754
           1       0.72      0.65      0.69      1246

    accuracy                           0.91      8000
   macro avg       0.83      0.80      0.82      8000
weighted avg       0.90      0.91      0.91      8000

plot_confusion_matrix(y_test, gs_rfc.best_estimator_.predict(X_test));

Visualising ROC_AUC curve

from scikitplot.metrics import plot_roc
y_pred_test = gs_rfc.best_estimator_.predict_proba(X_test)
plot_roc(y_test,y_pred_test);

Classifying real test data with the fitted model

y_pred = gs_rfc.best_estimator_.predict(test)

# Dataframe example
df = pd.DataFrame({'session_id':session_ids_for_test,'revenue':y_pred.ravel()})

# Save dataframe as csv file in the current folder
df.to_csv('hackathon_sunday_2.csv', index = False, encoding='utf-8') # False: not include index
df

	session_id	revenue
0	283149	0
1	253020	0
2	348435	0
3	299763	0
4	195120	0
...	...	...
9995	278347	1
9996	498240	0
9997	296710	0
9998	226309	0
9999	252349	0

10000 rows × 2 columns